Show HN: Trolling SMS spammers with Ollama
An experiment that uses a locally run large language model to text back SMS spammers has prompted debate over whether baiting scammers with bots is clever defense or wasted effort. Commenters question if the targets are real humans or automated systems, whether engaging them simply “warms up” numbers and helps evade carrier filters, and what legal risks arise if an AI appears to agree to real transactions. Others broaden the lens to an emerging AI arms race in messaging—spammers and recipients both deploying automation—and argue that regulatory enforcement, better identity protocols, or simply reporting spam may be more effective than elaborate counter-trolling.
Nature of the spammers
- Several commenters think initial messages are automated scripts, with humans stepping in only after a few “gates” are passed.
- Others argue some operations are almost fully automated and highly parallel, so tying up one “conversation” may not cost them much.
- There is debate about who the human operators are: claims range from kidnapped/forced workers to office-style scam shops in certain countries.
- Multiple people note that spammers already use LLMs and other automation tools.
Value and Risks of Trolling Spammers
- Supporters see value in diverting scammers’ attention from real victims and view this as a fun hobby project with minimal marginal cost.
- Critics argue that engaging at all may help “warm up” numbers and create “legitimate” traffic signals that improve spammers’ deliverability.
- Some recommend instead: reporting to carriers (e.g., 7726 in the US), using STOP, and leveraging newer text regulations and TCPA lawsuits.
- There is disagreement over how much the ongoing compute/server cost matters for such a hobby.
Technical Implementation Discussion
- Response latency is on the order of seconds for LLM plus extra time for the Android gateway.
- MQTT is praised as convenient but commenters note WebSockets, ZeroMQ, long-polling, raw TCP, VoIP/SIP trunks, or SBCs could all work.
- Ideas include adding random reply delays to seem more human, using Tailscale for connectivity, and replacing Android+SIM with cheap VoIP DIDs.
- SMS length limitations ( ~155 chars ) caused messages to be split in the demo.
Legal and Contract Concerns
- Several commenters worry about bots accidentally agreeing to real estate or car deals, potentially forming enforceable contracts via SMS.
- Others counter that contracts still require intent, consideration, and clear terms; automated replies may lack “meeting of minds.”
- Examples are raised where courts treated chatbot promises as binding on the company that deployed them, suggesting legal risk is non-trivial.
- Jurisdiction matters: some say phone-only agreements are weaker or require later written confirmation in parts of Europe.
Broader AI Arms Race & Alternatives
- Many find it darkly comic or depressing that LLMs will soon be talking mainly to other LLMs, wasting energy in an arms race of spam vs. counter-spam.
- Some argue this reflects “scarcity mindset” misusing technologies of abundance, and suggest more constructive uses like helping scammers find better work.
- Alternatives proposed include religious or moral outreach to scammers, better caller-ID / new phone protocols, and non-trolling applications (e.g., farmers interacting with LLMs via SMS).